zorkian / distributor Goto Github PK

It returns a response with content "file not found" and status 200 OK. It would be better to return a 404.

We are periodically seeing large bunches of
2016/11/28 20:04:05 [ERROR] Failed to bencode: write tcp X.X.X.X:6390->Y.Y.Y.Y:18014: write: broken pipe errors.

We'll see a lot of them over a period of a minute or two, and then no more. During this, there are also many open sockets in CLOSE_WAIT state. In some cases it recovers, but sometimes it runs out of fds and then crashes. It appears that this is somewhat load related, as it only shows up under heavy use. Over-aggressive retrying from ctorrent could definitely compound this (e.g.: if we fall behind and ctorrent retries timed out announces).

This seems to have started showing up after #17, so it might be due to the decreased Interval. It could also be due to the inefficient method of removing old peers. If we have a large peer list, we need to iterate over it on each announce. Previously, we only did so on new peers.

Possible solutions:

• Increase the interval back to the previous value, and likewise increase the timeout
• Only clean up old peers when we're attempting to hand them out

can you upgrade fsnotify? the project now lives in https://github.com/fsnotify/fsnotify